1. Field of the Invention
The present invention relates to emphasis processing for an image obtained by digital subtraction angiography.
2. Description of the Related Art
With recent advances in the digital technology, it has become popular to perform digital processing for images even in the medical field. Instead of conventional X-ray imaging using an X-ray diagnosis film, two-dimensional X-ray sensors configured to output an X-ray image as a digital image have also prevailed. Digital image processing such as edge emphasis processing for a digital image output from the two-dimensional X-ray sensor is important.
An application example of the digital image processing is DSA processing of acquiring a digital subtraction angiogram (to be referred to as a DSA image hereinafter). The DSA image is an image obtained by acquiring images before and after the inflow of a contrast medium into an object, and subtracting an image (to be referred to as a mask image hereinafter) before the inflow of the contrast medium from an image (to be referred to as a live image hereinafter) after the inflow of the contrast medium. In subtraction processing of the mask image from the live image, a blood vessel region serving as a region of interest in diagnosis is held as a change region between images that is generated by the inflow of the contrast medium. The remaining unnecessary region is removed as a background region, and the change region is expressed as a homogeneous region. The generated DSA image is an image helpful for diagnosis because the blood vessel region can be observed without influencing the remaining object structure.
The purpose of using a DSA image for diagnosis is clear visualization of a contrasted blood vessel image. This purpose is considered to have already been achieved in a subtraction image obtained by subtracting a mask image from a live image. However, an image more suitable for diagnosis is obtained by applying emphasis processing generally used as X-ray image processing to a subtraction image and further emphasizing the edge of a contrasted blood vessel region.
As an example of the emphasis processing, frequency emphasis processing will be explained. More specifically, an image is decomposed into a plurality of band-limited images representing different frequencies. The respective band-limited images undergo different adjustments and then are merged, generating one emphasized image.
As a method of decomposing an image into a plurality of band-limited images, there are various methods such as Laplacian pyramid decomposition, wavelet transform, and unsharp masking. For example, when unsharp masking is adopted, letting Sorg be an original image and Sus be a blurred image, a band-limited image H is given by:H(x,y)=Sorg(x,y)−Sus(x,y)  (1)where (x, y) is the pixel of an image, and H(x, y), Sorg(x, y), and Sus(x, y) are pixel values.
The above equation is used to generate one band-limited image from an original image. A method of generating a plurality of band-limited images representing different frequencies is as follows. A different frequency is represented by a level lv, and a plurality of band-limited images having different frequencies from a high frequency (lv=1) to a low frequency (lv=lvMax) are represented by {Hlv|lv=1, 2, . . . , lvMax}. At this time, the band-limited image Hlv at an arbitrary level lv is given by:Hlv(x,y)=Suslv-1(x,y)−Suslv(x,y)  (2)where {Suslv|lv=0, 1, 2, . . . , lvMax} are a plurality of blurred images having different frequencies. A blurred image Sus0 having lv=0 is the original image Sorg.
From equation (2), the relationship between the original image Sorg and the band-limited image Hlv is given by:Sorg(x,y)=Σlv=1lvMaxHlv(x,y)+SuslvMax(x,y)  (3)
This means that the original image Sorg can be reconstructed by adding all the decomposed band-limited images Hlv (to be referred to as high-frequency images hereinafter) and a blurred image SuslvMax having a lowest frequency (to be referred to as a low-frequency image hereinafter).
Based on this relationship, frequency emphasis processing is given by equation (4) below using a coefficient {αlv|lv=1, 2, . . . , lvMax} which gives an emphasis degree on a high-frequency image:Senh(x,y)=Σlv=1lvMaxαlvHlv(x,y)+SuslvMax(x,y)  (4)where Senh is an image having undergone frequency emphasis processing. When the emphasis coefficient αlv is set to be 1 at all levels, Senh becomes equal to the original image Sorg in accordance with equation (3). A high-frequency image is emphasized by setting the emphasis coefficient αlv to be larger than 1, and suppressed by setting it to be smaller than 1. That is, by setting a different value of the emphasis coefficient αlv for each frequency level lv, the user can create images of his preferences having undergone various frequency emphasis or suppression processes.
However, the emphasis coefficient αlv allows adjustment at each level, but emphasis or suppression is uniformly performed at the same frequency component. That is, this method has a problem that an edge component to be emphasized and a noise component to be suppressed cannot be separated.
To solve this problem, Japanese Patent Laid-Open No. 09-248291 discloses a method of detecting only an edge component from a high-frequency image and emphasizing it, thereby obtaining the emphasis effect of only the edge. Japanese Patent Publication No. 04-030786 discloses a method of performing threshold processing using a predetermined value for each portion of a subtraction image in order to clearly discriminate a blood vessel region in a DSA image from the remaining region, separating only the blood vessel region serving as a region of interest based on the result, and highlighting it.
The following problem arises when the above-described emphasis processing is applied to a DSA image.
The DSA image is a subtraction image obtained by subtracting a mask image from a live image and removing an object structure in order to enhance the contrast of a contrasted region.
In general, a small-pixel-value region in an X-ray image corresponds to a region having a thick object structure. Since the amount of X-rays reaching the sensor is small in the region having the thick object structure, the small-pixel-value region is a region where noise components with a low S/N ratio are dominant. Inter-image subtraction removes a clinically unnecessary object structure, but a noise component in such a small-pixel-value region remains on the subtraction image.
For this reason, if emphasis processing is applied to a DSA image, noise in a subtraction image corresponding to the small-pixel-value region before inter-image subtraction is further emphasized by emphasis processing, greatly impairing visibility.
The method disclosed in Japanese Patent Laid-Open No. 09-248291 in which only an edge component is detected and undergoes emphasis processing does not especially target a DSA image, and cannot perform emphasis processing considering noise in a subtraction image. The method disclosed in Japanese Patent Publication No. 04-030786 targets a DSA image, separates only a blood vessel region serving as a region of interest, and highlights it. However, this method does not consider noise dependent on a pixel value before the above-mentioned inter-image subtraction.